Song Byungdoo: "An Era Where Financial Institutions Are Chosen by AI, Not Customers" [2026 Financial Forum]

Customers Entrust Financial Strategies to AI, Not Apps

'Algorithm Selection' Emerges as New Competitive Edge for Financial Firms

Uncovering Previously Overlooked Innovative Companies and Individuals

The Best Capital Allocation Tool fo

Son Byungdoo, CEO of Toss Insight, offered insight that a new era is approaching in which artificial intelligence (AI) agents, rather than customers, will choose financial institutions. Instead of customers opening a bank application and comparing products themselves, it is highly likely that the mainstream approach will be for people to ask their AI agents to find personalized financial solutions. As a result, Son advises that financial institutions must go beyond simply improving algorithm performance and prioritize establishing company-wide data governance that clearly defines data ownership and responsibility in order to be chosen by AI agents.

"The Financial Paradigm Is Evolving from Product Sales to the 'Financial Planning & Execution Industry'"

Son Byungdoo, CEO of Toss Insight, is giving a lecture titled "AI Financial Innovation: The Power to Change the Flow of Capital" at the 2026 Asian Financial Forum held on May 21 at The Westin Josun Seoul in Jung-gu, Seoul, under the theme "Great Transition of Future Finance: The Era of Productive Capital and New Financial Order." 2026.5.21 Photo by Kim Hyunmin

On May 21, during his session titled "AI Financial Innovation: The Power to Change the Flow of Capital" at the 2026 Asian Financial Forum held at The Westin Josun Seoul in Jung-gu, Seoul, Son predicted, "Whereas finance in the past was structured around mass-producing and selling standardized products, the future of finance will evolve into a personalized financial planning and execution industry."

Son explained that leading global financial companies are already adopting AI as a core infrastructure, not simply as an option. According to a McKinsey survey, the annual economic value creation potential of generative AI is up to $4.4 trillion, which is on par with the GDP of the world’s fifth-largest economy. For example, JP Morgan has saved over 360,000 work hours annually by introducing an AI contract analysis system, while Bank of America's AI assistant "Erica" has surpassed 3 billion cumulative consultations. Goldman Sachs utilizes generative AI to assist with research drafts and risk analysis, and in the Chinese fintech sector, AI enables real-time loan screening to be completed in just a few minutes.

Son pointed out, "As AI becomes more powerful, the standard for competitiveness in financial institutions will shift from how many people they employ to how good their systems are." He particularly noted that as the primary customer interface shifts from app-based platforms to AI agents, the rules of financial competition will also be redefined. "In the past, marketing that directly persuaded customers was important, but in the future, being chosen by AI will become much more crucial," Son predicted. "The new source of competitiveness will not be brand power but rather which financial institution’s products are invoked and recommended by AI algorithms."

Shining a Light on Previously Unseen Companies and Individuals... Resolving Information Asymmetry

It was also suggested that the widespread adoption of AI finance could maximize the precision of capital allocation, thereby realizing both "productive finance" and "inclusive finance." Innovative companies have traditionally struggled to secure funding within existing financial systems due to insufficient financial statements and a high proportion of intangible assets such as intellectual property (IP) and patents. However, Son explained that if AI can comprehensively analyze unstructured data like transaction networks, sales patterns, and industry trends, it would be possible to identify promising companies with high growth potential at an early stage. He also noted that personal finance will become more inclusive, further bringing marginalized groups into the mainstream financial system.

Referring to a recent comment by the policy chief at the Office of the President on social media that "finance has been cruel," Son strongly emphasized the need to lower the barriers of the traditional financial system. When evaluating the credit of "thin filers" with limited transaction history, alternative credit assessments become possible by having AI comprehensively analyze data such as telecommunications and gas bill payment records, social insurance subscription details, and platform transaction data, instead of relying solely on traditional credit scores. Furthermore, real-time data analysis enables early detection of risk signals before delinquencies occur, making it possible to establish a "preventive finance system" linked with community financial institutions.

Son explained, "AI is a technology that makes previously unseen customers and companies much more visible," adding, "AI’s greatest significance in productive finance is its ability to precisely determine where capital should flow and thereby enhance the accuracy of capital allocation."

Korean-Style "Departmental Silos" Are the Biggest Obstacle... Regulation Should Also Emphasize 'Outcome Responsibility'

However, Son identified "five structural barriers," including Korea's unique siloed organizational culture, that make it difficult for AI to take root in the domestic financial industry. These include: (1) siloed organizations that cause fragmentation and ownership conflicts over data across departments and affiliates; (2) legacy systems centered around 20- to 30-year-old mainframes; (3) data quality issues with inconsistent formats and frequent errors; (4) regulatory and compliance risks where the cost of accidents is believed to outweigh the benefits of innovation; and (5) a conservative decision-making culture that hesitates to pursue innovation due to unclear responsibility for AI failures.

Son particularly criticized the issue of data quality. He argued that despite domestic financial institutions spending vast amounts to adopt AI, more than half of project time is wasted on disparate "data cleansing" tasks rather than developing advanced models, due to the lack of standardization. While data cleansing is essential even if time-consuming, in reality, formats differ between departments, duplication and errors persist, and standardization is lacking. Therefore, he believes that establishing robust data governance should be the top priority.

He noted, "What matters for AI is not the quantity of data, but its quality," adding, "Some financial institutions report that over half of project time is spent just on cleaning up data." He continued, "Financial institutions must establish clear data governance before worrying about algorithms. Traditional financial firms face difficulties adopting AI not because of a lack of technology, but due to rigid organizational structures, poor quality data, and conservative culture."

Regulatory Paradigm Must Also Shift to a 'Negative' Approach

Lastly, Son strongly advocated for a shift in the regulatory paradigm by financial authorities. Under the current "positive regulation" system, where everything not explicitly permitted is prohibited, new services require regulatory approval each time, making it impossible for the market to keep pace with the speed of learning.

Son argued, "Future AI finance regulation should not be about 'what not to do,' but should shift to strengthening outcome responsibility and principle-based regulation, focused on 'what to be responsible for.'" He emphasized that innovation can only occur if companies that faithfully comply with minimal procedural standards are rewarded with exemption from liability for uncertain outcomes.

He also predicted, "The subject of regulation will shift from people (employees) to algorithms (designers, data flows)," and "the paradigm will evolve from the traditional financial supervision model of monitoring ‘what was explained’ to a log-based real-time supervision system that verifies what data the AI viewed and which paths it used to make recommendations, along with function-based regulation."

This content was produced with the assistance of AI translation services.